CPU Power Consumption & Battery Life Calculator

Introduction to estimating laptop runtime from CPU and platform power

This calculator estimates how long a laptop runs on battery by dividing the pack's usable stored energy by the average power the whole system pulls out of that pack. Both halves of that division are easy to get wrong, and this page is built around getting them right rather than around producing a comfortable-looking number.

The energy half is a unit problem. A battery's stored energy is an energy quantity measured in watt-hours (Wh). Manufacturers frequently publish a charge quantity instead — milliamp-hours (mAh) — and mAh on its own is not energy. IEC 61960-3:2017 defines rated capacity as a quantity of electricity in ampere-hours and separately defines a nominal voltage, and its marking clause requires a cell or battery to carry both the rated capacity and the nominal voltage. The Smart Battery Data Specification takes the same view at the firmware level: DesignCapacity() reports either mAh or 10 mWh depending on a mode bit, and DesignVoltage() is a separate register in millivolts. So this calculator refuses to treat a bare mAh figure as energy. Choose mAh and it will ask for the pack's nominal voltage, then convert.

The power half is a systems problem. The CPU is not the whole laptop. In the ENERGY STAR Version 8.0 sample calculation for a notebook computer, the measured whole-system power is 10.0 W in Short Idle (screen on, machine doing nothing) and 6.0 W in Long Idle (backlight off). A modern 15 W-class CPU package idles at roughly one to two watts, so at idle the display, DRAM, SSD, wireless radios, embedded controller, fans and voltage-regulator losses account for the large majority of the drain. Any page that sets system draw equal to CPU TDP will be badly wrong at exactly the workloads where people care most about battery life.

The third trap is TDP itself. Thermal Design Power is a cooling-design figure, not a consumption figure, and the two vendors define their numbers differently. Intel's client datasheets equate Processor Base Power (a.k.a. TDP) with the recommended setting for PL1, "a threshold for average power that will not exceed", while PL2 permits a much higher short burst governed by an averaging constant called Tau — an Intel 45 W H-line mobile part is specified with PL1 = 45 W and PL2 = 115 W. Intel states plainly that "Processor Base Power is not the maximum power that the processor can dissipate". AMD is equally direct in its own power-management documentation: some parts "are rated at a 400W TDP even though they will rarely ever draw this amount of power", because the rating exists to size platform current, voltage and cooling. This calculator therefore asks for a sustained package power limit (PL1, cTDP or PPT) and treats it as the ceiling of an active-power range, not as a constant draw.

How to use the inputs: pack energy first, then the power budget

Work outward from the battery. Every input has a realistic default, so you can change one field at a time and watch the runtime, the power breakdown chart and the load-sensitivity table respond.

1. Pick the capacity unit. If the label, spec sheet or OS battery report gives watt-hours, choose Wh and enter it. If it gives mAh, choose mAh and the pack's nominal voltage field becomes required — a three-cell Li-ion pack is normally 11.1–11.55 V and a four-cell pack 14.4–15.4 V, because IEC 61960-3 puts a single Li-ion cell at a 3.7 V nominal and a series string multiplies that.
2. Set battery health. This is FullChargeCapacity() divided by DesignCapacity() — exactly the ratio the Smart Battery Data Specification says a host uses "to determine battery wear". Windows powercfg /batteryreport and macOS System Information both surface it.
3. Set the usable discharge window. Firmware and the OS stop the machine before the pack is genuinely empty, so usable energy is always less than full-charge energy. 92–97% is a reasonable band; 100% models an idealised pack.
4. Enter the sustained CPU package power limit. Use PL1 (Intel), cTDP, or PPT (AMD) — the number the machine will actually hold under a long all-core load, which on thin laptops is often configured well below the marketing TDP.
5. Enter CPU package power at idle. One to three watts is typical for a modern mobile SoC. This is the floor the model never drops below, which is what stops the "idle CPU draws zero watts" error.
6. Choose a CPU utilisation preset. The five steps mirror the way SPECpower_ssj2008 characterises a machine — it sweeps target loads "from 100% to 0% in increments of 10%" plus an Active Idle point — because power is only meaningful when you say which load level you mean.
7. Enter GPU power. Leave it at 0 W when the discrete GPU is parked and the integrated GPU's power is already inside the package limit above; raise it when a dGPU is awake.
8. Enter display power at full brightness and the brightness you actually run. The ENERGY STAR notebook sample gives a useful sanity check: turning the backlight off dropped that machine from 10.0 W to 6.0 W.
9. Enter other platform power — DRAM, SSD, Wi-Fi, Bluetooth, embedded controller, fans, USB and Thunderbolt retimers. This is the bucket most people under-estimate.
10. Set the DC-DC conversion efficiency. The rails that feed the CPU, DRAM and display are all fed from the pack through regulators, and their losses come out of the battery too.
11. Optionally add an undervolt in millivolts together with the nominal core voltage it is applied to, then press Calculate Battery Life. Download Power Report saves the whole scenario, including the sensitivity sweep, as CSV.

The runtime formula, the unit chain and the PL1 power model

Everything on this page reduces to one division — energy divided by power — but each side of that division is built up carefully.

Step 1: get the pack into watt-hours

Charge in milliamp-hours only becomes energy once it is multiplied by the voltage at which that charge is delivered. For a pack quoted in mAh:

A 5000 mAh pack is 18.5 Wh at 3.7 V and 55.5 Wh at 11.1 V. Those are different batteries by a factor of three, which is exactly why a calculator that accepts mAh without a voltage is not doing physics.

Step 2: rated energy is not usable energy

A pack loses capacity with age and cycling, and the system never discharges it to zero. Both effects are multiplicative:

is the wear ratio named in the Smart Battery Data Specification, and f_window is the fraction of the full charge the platform will actually spend before it shuts down.

Step 3: CPU package power from a sustained limit, not from TDP alone

The model treats the sustained limit (PL1, cTDP or PPT) as the ceiling of an active range that starts at the measured idle package power, and interpolates linearly with utilisation u:

At u = 0 the CPU still costs you its idle power; at u = 1 it sits at the sustained limit. Neither end is TDP-as-constant-draw. Note what this deliberately does not model: PL2 bursts. Intel's Tau lets a package exceed PL1 for tens of seconds, so a bursty workload can consume more energy than a flat PL1 model predicts even though its average utilisation looks low.

Step 4: undervolting scales the dynamic term quadratically

Switching power in CMOS logic goes with the square of supply voltage. The page applies that square law to the active part of package power only, because the idle floor is dominated by leakage and always-on domains that a core undervolt barely touches:

A 100 mV undervolt on a 1.05 V rail gives k = (0.95/1.05)² ≈ 0.818, so about 18% off the active term. The previous version of this page used a linear rule that worked out to 0.3% for the same 100 mV — a hundredfold understatement — which is one of the defects fixed in this revision.

Step 5: display, platform and conversion losses

Display power is modelled as a fixed panel-electronics share plus a backlight share that tracks brightness linearly. The 15% floor below is a stated modelling assumption of this page, chosen so that a panel at minimum brightness still costs something; it is not a figure taken from any of the standards cited below.

Adding the rest of the platform and dividing by the regulator efficiency η gives the current actually drawn out of the cells:

Step 6: the runtime division

Because power sits in the denominator, runtime is a hyperbola, not a line. Going from 10 W to 20 W costs you half your runtime; going from 50 W to 60 W costs about a sixth of it. The load-sensitivity table the calculator prints makes that curvature explicit, and it is the single most useful thing to look at before deciding which knob to turn.

Worked example: a two-year-old 28 W ultrabook doing office work

These are the values the form is pre-loaded with, so you can press Calculate Battery Life and check every line below against the output.

The label on the pack says 5000 mAh, 11.4 V. In watt-hours that is 5000 × 11.4 / 1000 = 57.0 Wh — enter either form and the calculator lands in the same place. The battery report says the pack now holds 90% of its design capacity, and the platform reserves the bottom of the discharge curve, so the usable window is 95%.

• Usable energy = 57.0 × 0.90 × 0.95 = 48.74 Wh (already 8.3 Wh less than the number on the sticker)
• CPU package = 1.5 + (28 − 1.5) × 0.25 = 8.13 W at the 25% Light preset
• GPU = 0 W, because the dGPU is parked
• Display = 4.5 × (0.15 + 0.85 × 0.60) = 2.97 W at 60% brightness
• Other platform (DRAM, SSD, Wi-Fi, EC, fans) = 4.00 W
• Component total = 8.13 + 0 + 2.97 + 4.00 = 15.10 W
• Battery-side draw = 15.10 / 0.87 = 17.35 W
• Runtime = 48.74 / 17.35 = 2.81 h = 2 h 49 min

Two things in that chain are worth pausing on. First, the CPU contributes 8.13 W of a 17.35 W battery-side draw — a bit under half, at a load level most people would call "light office work". Take the utilisation preset down to Idle and the CPU falls to 4.15 W while the display, platform and regulator losses barely move; runtime rises to 3 h 49 min rather than to some idealised all-day figure, because you cannot switch off the other 60% of the budget. Second, ageing and the discharge window between them removed 14.5% of the pack before a single watt was drawn. A calculator that divides the sticker capacity by the load will always be optimistic on a real, used machine.

Now change one input at a time from the baseline. Drop brightness from 60% to 20% and display power falls from 2.97 W to 1.44 W, worth about 19 extra minutes. Apply a 100 mV undervolt at a 1.05 V core and the active CPU term drops by 18%, worth about 15 minutes — a genuinely comparable lever at this load. Wake a 25 W discrete GPU, though, and battery-side draw jumps from 17.35 W to 46.09 W and runtime collapses to just over an hour: one decision that outweighs every other setting on the page combined. Ranking the levers correctly is the entire point of the exercise, and the ranking changes with the workload.

Interpreting the result: reading whole-system watts, not CPU watts

The headline number is a projection for a steady workload held for the whole discharge. Read it as a budget rather than a prophecy: it tells you how many watts you can afford, and the breakdown chart tells you where those watts are going. The table below anchors the scale using published whole-system measurements rather than guesses. The 6.0 W and 10.0 W rows are the measured values in the ENERGY STAR Version 8.0 sample calculation for a notebook computer; the rest are typical operating bands for a thin-and-light class machine. Runtimes assume the worked example's 48.7 Wh of usable energy.

Three readings follow from that shape. If your result sits near the top of the table, the CPU inputs barely matter and you should spend your attention on the display and the platform bucket. If it sits in the middle, the CPU and the display are trading places and the load-sensitivity sweep is the fastest way to see which. If it sits at the bottom, nothing except the GPU decision will move the needle, and no amount of brightness tuning will rescue a two-hour session.

It is also worth checking the result against your own experience, because that calibrates the inputs you are least sure of. If a 57 Wh pack at 90% health really lasts six hours of light browsing, your actual average battery-side draw is 48.7 / 6 ≈ 8.1 W — below the ENERGY STAR Short Idle figure for that sample machine, which tells you your display power or platform bucket is set too high in the form.

Limitations, assumptions and where this model stops being trustworthy

The model is deliberately a steady-state energy balance. It is honest about the unit chain and about the fact that a CPU is one consumer among many, but there are several real effects it does not capture, and you should know which they are before you rely on the number.

• PL2 and boost bursts are not modelled. Intel's package power control lets average power exceed PL1 for a window governed by Tau — an H-line 45 W part is specified with PL2 = 115 W — and AMD's SMU makes equivalent opportunistic decisions inside PPT. A workload made of many short bursts can therefore draw more energy than a flat interpolation from PL1 suggests.
• Utilisation is not proportional to power. An AVX-heavy loop and an integer loop at the same reported "50%" draw very different package power, which is precisely why SPECpower_ssj2008 defines a calibrated target-load sweep instead of trusting an OS utilisation figure.
• mAh capacity is rate-dependent. IEC 61960-3 defines rated capacity as the ampere-hours delivered over a five-hour discharge, and the Smart Battery Data Specification qualifies its mAh figures "at a C/5 discharge rate". Discharge a pack much faster than C/5 and you get fewer usable amp-hours than the label implies, so heavy-load runtimes are optimistic by a little more than the arithmetic shows.
• Voltage sags as the pack empties. Nominal voltage is, in IEC's own words, a "suitable approximate value of the voltage used to designate or identify a cell". A Li-ion cell nominally at 3.7 V is discharged down to about 2.5 V, so instantaneous Wh conversion drifts through the discharge even though the total energy figure remains a fair average.
• The 15% display floor and the linear brightness term are page assumptions. They are not drawn from any standard. Panel technology matters enormously here: an OLED's power depends on displayed content, not just on the backlight setting, and this model has no way to represent that.
• Conversion efficiency is treated as a single constant. Real buck converters are less efficient at very light loads, so a laptop idling at 5 W typically converts worse than the same laptop at 25 W. Assuming one efficiency across the whole range flatters idle scenarios.
• Capacity fade is a single number here. State of health depends on cycle count, calendar age, storage temperature and how often the pack sat at 100%; a single percentage cannot express any of that, and it changes while you own the machine.
• Undervolt stability is not guaranteed. Many platforms disable core voltage offsets in firmware, and an aggressive offset causes silent corruption or crashes long before it causes an obvious failure. Treat the quadratic term as an upper bound on the benefit.
• Nothing here models the operating system's own behaviour — background indexing, sync, wake locks, a Wi-Fi radio hunting for a weak access point, or thermal throttling that changes the power/performance point mid-session.

Input guidance: realistic ranges and how to calibrate the model

If you are unsure what to enter, start with a plausible band and then calibrate against one runtime you have actually observed. Precision in one measured anchor beats guessing at eight fields.

Typical bands by machine class

These are operating bands, not specifications. Vendor firmware, panel size, refresh rate and cooling design move them substantially, and the same CPU part is routinely configured to different sustained limits in different chassis.

• Thin-and-light: sustained package limit 9–28 W (Intel's U-line is specified at PL1 = 9 W or 15 W, the P-line at 28 W), idle package 1–2 W, GPU 0 W, display 2–5 W at full brightness, other platform 3–6 W.
• Mainstream 14–16 inch: sustained package limit 28–45 W, idle package 1.5–3 W, GPU 0–30 W depending on whether the dGPU is awake, display 4–8 W, other platform 4–9 W.
• Gaming or mobile workstation: sustained package limit 45–65 W (Intel's H-line is specified at PL1 = 45 W with PL2 = 115 W), idle package 2–4 W, GPU 10–175 W, display 6–12 W for a high-refresh panel, other platform 6–15 W.

A calibration procedure that actually converges

Measure one runtime under a workload you can repeat — a full discharge while browsing, with brightness fixed. Multiply your rated Wh by your health percentage and by the usable window to get usable energy, then divide by the hours you observed. That quotient is your true average battery-side draw.

1. Set the utilisation preset to match what you were doing (Idle or Light for browsing).
2. Set the sustained package limit and idle package power from your CPU's specification.
3. Set display power from your panel's rating, or estimate it as the ENERGY STAR Short Idle minus Long Idle difference for a similar machine — 4.0 W in the published sample.
4. Adjust the other platform power field, which is the one you know least, until the calculated runtime matches your measurement.

Once the model reproduces one real observation, every comparison it makes afterwards is grounded. Changing brightness, waking the GPU or raising the load then produces differences you can trust, because the baseline is no longer a guess.

Frequently asked questions about CPU power, TDP and battery runtime

Sources and standards used for these definitions

Power-limit definitions: Intel Corporation, 12th Generation Intel® Core™ Processors Datasheet, Volume 1 of 2 (document ID 655258), sections "Package Power Control" and "Processor Line Thermal and Power Specifications" — edc.intel.com. AMD, AMD EPYC™ 8004 and 9004 Series CPU Power Management white paper — amd.com. Battery capacity rating: IEC 61960-3:2017, Secondary cells and batteries containing alkaline or other non-acid electrolytes — Secondary lithium cells and batteries for portable applications — Part 3, clauses 3.4 nominal voltage, 3.5 rated capacity and 5.2 marking — webstore.iec.ch. SBS Implementers Forum, Smart Battery Data Specification, Revision 1.1, sections 5.1.17, 5.1.23 and 5.1.24 — sbs-forum.org. System-level power measurement: US EPA, ENERGY STAR Program Requirements Product Specification for Computers, Version 8.0, operational-mode definitions, Table 5 notebook mode weightings and the notebook sample calculation (Off 0.5 W, Sleep 1.0 W, Long Idle 6.0 W, Short Idle 10.0 W), together with the ENERGY STAR Test Method for Computers, which implements IEC 62623 Ed. 1.0 — energystar.gov. Workload load-level definitions: Standard Performance Evaluation Corporation, SPECpower_ssj2008 Design Document, section 2.1 "Load Levels" — spec.org. The 15% display-power floor, the linear brightness term and the linear utilisation interpolation are modelling assumptions of this page and are not taken from any of the documents above.